Models

ModelService (dodil.ignite.v1.ModelService) provides managed inference over a global model catalog. It is OpenAI/Cohere-compatible: chat completions, embeddings, rerank, and transcription, plus a generic Infer.

A ModelInfo’s declared input/output format routes you to the right RPC — read the catalog first to find a model and the surface it supports.

Powered by Ignite. Models isn’t a separate product — it’s consolidated onto the same Ignite infrastructure that runs your apps, surfaced as its own OpenAI/Cohere-compatible API (independent of any app you deploy). Inference is offered serverless through this catalog today; running your own model code on GPU-backed apps (via Compute) is on the roadmap — until then, Models is how you run inference on Ignite.

In this section

• Model Catalog — supported models, their specs, and how each is billed (per token or per compute time)
• Using OpenAI & Cohere SDKs — point the official SDKs at Ignite, no Dodil SDK needed
• API Reference — every ModelService RPC, with OpenAI/Cohere-style HTTP paths

CLI: dodil ignite models — list, get, chat, embed, infer, transcribe (rerank + streaming variants are API-only). See CLI Basics.

See also

• API Reference
• Conventions — OpenAI-style /v1/... paths, google.protobuf.Value polymorphic fields, streaming
• Auth and Access